Title: Japanese Mora-Timing and Speech Processing: The Case of Devoiced Vowels
1Japanese Mora-Timing and Speech Processing The
Case of Devoiced Vowels
- Natasha Warner and
- Naomi Ogasawara
- University of Arizona
2Background on Mora-Timing
- Japanese moras are often said to be of
approximately equal duration - /tookyoo/ /karaoke/
- µµ µ µ µ µµ µ
- Vowel devoicing is one influence against
regular timing devoiced vowels are shorter. - /kita/ ki8ta North /kusa/ ku8sa grass
- /mita/ mita saw /kuma/ kuma bear
3Perception and processing
- Most devoicing and mora-timing literature is on
production. - Kato duration changes, even compensatory ones,
make speech less acceptable to listeners. - If regular timing of moras is very important,
listeners may use it in processing speech. - When moras are not regular because of devoicing,
this could make processing difficult. - But if listeners use the phonology, they may
expect devoicing exactly where it happens.
4Processing of devoiced vowels
- Phoneme monitoring (non-words), with voiced and
devoiced vowels both in appropriate and
inappropriate environments - Listeners hear hoki8to or hokito
tadZiga or tadZi8ga - Press the button as fast as you can whenever you
hear i as in i, ki, Si, - Lexical decision real words, press the button
if the thing you hear is a real word - Example words aki?kan or akikan empty
can gend?itsu or gend?i?tsureality
5Examples of stimuli
- Phon. monitoring Lex. decision
- hoki8to gend?i?tsu
- hokito gend?itsu
Does target mora duration differ in the stimuli?
6Yes, devoicing shortens stimuli
- Target mora duration and total item duration are
significantly shorter for devoiced vowels
mora dur. (ms)
(non-)word dur. (ms)
Lex. Decision
aki?kan akikan gend?i?tsu gend?itsu
aki?kan akikan gend?i?tsu gend?itsu
Phon. Monit.
hoki?to hokito tad?i?ga tad?iga
hoki?to hokito tad?i?ga tad?iga
7Effect on processing words
- Are devoiced stimuli harder to process because
they diverge from mora-timing? - Reaction Times (ms)
Error Rates ()
aki?kan akikan gend?i?tsu gend?itsu
aki?kan akikan gend?i?tsu gend?itsu
No. Phonologically appropriate word forms are
easier to process (Lexical decision task).
8Effect on processing sounds
- Phoneme monitoring task Reaction times (ms)
Voicing env. appropriate form is easier to
process.
Devoicing env. No effect. Devoiced i8 is
often deleted, making the phoneme hard to
recognize. In the devoicing environment, this
balances out the preference for the
phonologically appropriate form.
hoki?to hokito tad?i?ga tad?iga
9Intermediate summary
- Vowels that are appropriately voiced or
appropriately devoiced are easier to process than
those that are inappropriate. - Listeners prefer phonologically appropriate
forms, even if that means a vowel is devoiced,
and a mora is shorter. - ? Phonology outweighs any preference for moras to
be regularly timed.
10A second test splicing
- Control perception experiments target mora
spliced into surrounding (non-)word stimulus, so
that the only thing that differs between
conditions is the /i/-mora itself.
h o k j t o
h o k i t
o
Devoiced moras are significantly shorter.
11Splicing hypothesis
- Meant as a control, but provides an interesting
test of processing of timing compensation. - Port et al. if one mora is short, others in the
word lengthen to compensate. Duration of word
remains predictable. - Splicing disturbs any compensation relationship.
- If there is word-level compensation, spliced
stimuli should be harder to process, because they
violate timing patterns.
12Splicing perception results
Reaction Times (ms) Error
Rates ()
Lex. Decision Phon. Monit.
Spliced stimuli are no harder to process.
13Conclusions
- Listeners find it easier to process forms with
phonologically appropriate (de)voicing, even
though moras diverge from regular timing. - Altering the compensation relationship within the
word does not hinder processing. - Listeners are sensitive to the phonological
patterns in their language, and use them more
than any expectation of regular timing in
processing.